AMD Mem Tweak XL - Read/modify timings/pp/straps on the fly

[PharmEcis]

What I'm able to decipher here is that someone got a just desert served nice and cold and can no longer steal another's glory like it was done last time.  

KUDOS on every single fucking front mate!

Time to convert my vega boxes to linux!

[1714246649]

1714246649

[Docice]

Hi, does it work with rx580?

[miner29]

Looks like it works with everything team RED.

Just getting to play with it but very NICE.
Thanks to Eliovp for sharing with us little peeps!

[nordmann666]

pls...step by step guide for hiveos

and how much more on CN?

[Bazzaar]

lol! all the useless leeches bleating "Gimme,Gimme! I'm incapable of actually doing anything difficult and need spoon feeding!"

[dagarair]

OhGodAnHELIOXlargementPill
  (theoriginalgangsta)
 

[highaltitude]

How do you apply this .cpp file?

[scryptr]

OhGodAnHELIOXlargementPill
  (theoriginalgangsta)
 

THE EXECUTABLE IS NOT LIKE THE PILL--

The ETHlargementpill is "full auto", it runs for the 1080ti and is pre-programmed with memory timings.  However, the AMDmemtool is very basic and allows the user to apply each memory tweak separately.  The user must put his own tweaks in place, and understand the terms and shorthand for each tweak.

This tool is usefull for a user to put together custom timings prior to programming the GPU BIOS with straps.  An analogy: like giving a user a stickshift assembly, but requiring the user to bolt it into the manual transmission case and mounting it in the car himself.

The "*.cpp" file compiles readily if the instructions in the README.md are followed.  It reads data from my R9 Fury cards with no problem.  But, as I do not know the tweaks necessary to improve the "--current" settings, I need to refer to a memory timings manual.  The downloadable binary from the Releases page works the same as my own compiled binary.       --scryptr

[ocminer]

Expect this tool, maybe super-slightly optimized closed source and with 2-5% dev-fee under some other name(s) soon

[Eliovp]

@OP, can you do me a favor?

Can you cat /sys/kernel/debug/dri/%d/amdgpu_regs for me please? is amdgpu_regs static, or does it change?

Have you tried it yourself? 

nice work, taking mining back to the old days of people helping each other, and not being a bunch of greedy bastards! 

Thank you

What I'm able to decipher here is that someone got a just desert served nice and cold and can no longer steal another's glory like it was done last time.  

KUDOS on every single fucking front mate!

Time to convert my vega boxes to linux!

Very welcome!

Hi, does it work with rx580?

It sure does.

OhGodAnHELIOXlargementPill
  (theoriginalgangsta)
 

THE EXECUTABLE IS NOT LIKE THE PILL--

The ETHlargementpill is "full auto", it runs for the 1080ti and is pre-programmed with memory timings.  However, the AMDmemtool is very basic and allows the user to apply each memory tweak separately.  The user must put his own tweaks in place, and understand the terms and shorthand for each tweak.

This tool is usefull for a user to put together custom timings prior to programming the GPU BIOS with straps.  An analogy: like giving a user a stickshift assembly, but requiring the user to bolt it into the manual transmission case and mounting it in the car himself.

The "*.cpp" file compiles readily if the instructions in the README.md are followed.  It reads data from my R9 Fury cards with no problem.  But, as I do not know the tweaks necessary to improve the "--current" settings, I need to refer to a memory timings manual.  The downloadable binary from the Releases page works the same as my own compiled binary.       --scryptr

No, true, it's not like "the pill".
It's better.
- You have the option to finetune yourself.
- It's open source
- It's based on data discovered by myself and not someone else. The only person that helped me to write the code is mentioned.
- ....

So no, it's not the same no

Btw, you can run the tool with multiple flags, ./amdmemorytweak --i 0,2,5 --CL 19 --RFC 260 --RAS  ... etc

Expect this tool, maybe super-slightly optimized closed source and with 2-5% dev-fee under some other name(s) soon

There's no "optimized" source possibility here. Exactly as Laik pointed out.

The only thing that could be done is automate the process with predefined timings. Something you can do yourself.
Heck, you can do even better if you take the time to figure them out.

Reading your comment makes me extra happy that i released this open source because 2-5% is in my eyes "theft".

Cheers!

[Remember remember the 5th of November]

So I managed to get this working on Windows. Are the values correct for r9 380x?

https://imgur.com/a/hZ7btRa

[Remember remember the 5th of November]

So I managed to get this working on Windows. Are the values correct for r9 380x?

https://imgur.com/a/hZ7btRa

No, R9 have different struct.

Are these readily available?

[Remember remember the 5th of November]

So I managed to get this working on Windows. Are the values correct for r9 380x?

https://imgur.com/a/hZ7btRa

No, R9 have different struct.

Are these readily available?

https://raw.githubusercontent.com/ddobreff/amdmeminfo/master/gmc.h

       ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_wr_ctl_d0), pos );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_wr_ctl_d1), pos+8 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_ras_timing), pos+16 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_cas_timing), pos+24 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_misc_timing), pos+32 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_misc_timing2), pos+40 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_pmg_timing), pos+48 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_misc1), pos+56 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_misc3), pos+64 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_seq_misc8), pos+72 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_arb_dram_timing), pos+80 );
        ultostr16_8( uint32_swap2(*(uint32_t *)&mc_arb_dram_timing2), pos+88 );

And this is their order.

Thanks!

[pbfarmer]

do you happen to know if there's an addressable edc sensor/counter on vega?  I remember someone (nerdralph?) talking the possibility a year back, but never really saw anything come of it.  otherwise, does syslog capture all hw faults, or only in some cases?  i've seen them reported in the past, but was not convinced the reporting was comprehensive.

would be useful in whipping up a nice little auto-tuner.

[ustcstone]

I successful managed to run the tool on hiveos(untubu 14.04),but i have no idea how to use it......
I try ./amdmemorytweak --i 0 -CL 20  (default is 21),but gpu0 got offline after a second.....
a little more guidance in the readme will be grateful!

[Docice]

And how to use? Are there optimal settings for the rx580?

[rdome]

For rx580 this thing got no use, my opinion. You have allredy stripped firmwares all over the net, why you would use this for 580
This has use for Vega cards because you dont have other option

[heavyarms1912]

Hey! it's you.

Great work and thanks for sharing.

[heavyarms1912]

For rx580 this thing got no use, my opinion. You have allredy stripped firmwares all over the net, why you would use this for 580
This has use for Vega cards because you dont have other option

So that you can experiment with live values instead of BIOS modding every time?

[windward]

can someone please share the config that they used for their vega and can someone explain or just give full forms of every parameter